JP2003508806A - Transmission system with improved encoder and decoder - Google Patents

Abstract

(57) Abstract In sine audio encoders it is known to use different time bases to analyze different parts of the frequency spectrum. In conventional encoders, sub-band filtering is used to divide the input signal into a number of sub-bands. Dividing the input signal into sub-bands can result in signal components at the boundaries of the two bands being represented in the signals of both sub-bands. This double representation of the signal components can lead to some problems in encoding these components. According to the invention, it is proposed to use prevention means (46, 48, 58, 68; 88, 92, 96) which prevent the signal components from having multiple representations.

Description

Detailed Description of the Invention

[0001]

【Technical field】

The present invention relates to a transmission system comprising a transmitter comprising an audio encoder, said audio encoder comprising segmenting means for deriving at least a first signal segment and a second signal segment from an input signal representing an audio signal,
The first signal segment is longer than the second signal segment, the audio encoder has means for deriving an encoded audio signal from the first and second signal segments, and the transmitter is via a transmission medium. The receiver has transmitting means for transmitting the encoded audio signal, the receiver has receiving means for receiving the encoded audio signal from the transmission medium, and the encoded means It relates to a transmission system further comprising an audio decoder for deriving a decoded audio signal from the audio signal.

The invention also relates to a transmitter, an encoder, a coding method, a tangible medium carrying a computer program for carrying out the coding method, and a signal carrying the computer program for carrying out the coding method.

[0003]

[Background technology]

A transmission system as described in the preamble of claim 1 is disclosed in US Pat. No. 5,886,276.
It is known from the issue.

Such a transmission system and an audio encoder store an audio signal in a storage medium which must be transmitted via a transmission medium having a limited transmission capacity or has a limited storage capacity. Used in applications where it must be. Examples of such applications include the transmission of audio signals over the Internet, the transmission of audio signals from mobile phones to base stations and vice versa, and the storage of audio signals in CD-ROM, solid state memory or hard disk drives. There is.

In order to achieve good audio quality at limited bit rates, audio encoders of different working principles have been tried. In one of these operating methods, the audio signal to be transmitted is divided into a plurality of segments, which typically have a fixed length of 10 to 20 ms. In each of these segments, the audio signal is represented by a plurality of signal components, which may be sinusoids defined by amplitude, frequency and possibly phase.

The transmitter sends the amplitude and frequency representations of the signal components to the receiver. The operations performed by the transmitter may include channel coding, interleaving and modulation.

The receiving means receives a signal representing the audio signal from the transmission channel and performs operations such as demodulation, deinterleaving and channel decoding. A decoder derives a reproduced audio signal from the representation by receiving the representation of the audio signal from the receiving means, generating a plurality of sinusoids represented by the encoded signal and combining them into an output signal. To do.

The problem with these audio encoders lies in choosing the proper length (in time units) for the signal segment. Long signal segments make it possible to obtain good frequency resolution for the determination of signal components, but a phenomenon called pre-echo can occur as a result of the limited temporal resolution. Pre-echo occurs when a sudden attack-like event of an audio signal is already heard before the actual occurrence of such event. The problem with pre-echo does not occur when the signal segment is short, but the frequency resolution for the determination of signal components with low frequencies is dramatically reduced.

In order to improve this, in said US patent the input signal is divided into a number of sub-bands by means of sub-band filters and a signal segment of different length is selected for each of these sub-bands. The lengths of these signal segments are chosen to be inversely proportional to the frequency range of the corresponding subband.

The problem with this method is that the coding quality for signal components located near the transition band of the sub-band is lower than that for other signal components.

[0011]

DISCLOSURE OF THE INVENTION

An object of the present invention is to provide a transmission system as described in the above-mentioned front part, which solves the above-mentioned problems.

In order to achieve the above object, in the transmission system according to the present invention, the encoding means includes:
It is characterized in that it has a prevention means for preventing multiple representations of a single signal component from occurring in the encoded audio signal.

The invention is based on the recognition that in conventional systems the frequencies in the transition band of the subband filter lead to multiple representations of the same signal component in the input signal. These multiple representations are not preferred where a psychoacoustic model is used to determine the signal components to be transmitted. Moreover, it is difficult to reconstruct the signal components that are represented twice in the coded signal. Also, multiple representations lead to higher bit rates than would exist if there were no multiple representations of signal components.

By using the above prevention measures to prevent or suppress these multiple representations of a single signal component, the associated problems will also be eliminated.

In one embodiment of the invention, the preventing means comprises a combining means for deriving a combined audio signal from a portion of the encoded audio signal representing the first signal segment and a signal representing the input signal. Subtracting means for deriving the second signal segment by subtracting the synthesized audio signal. The signal component determined from the first signal segment is removed from the signal representing the audio signal by subtracting the composite audio signal representing the first signal segment from the signal representing the audio signal to obtain the second signal segment. Like As a result, these signal components are either absent or strongly attenuated in the second signal segment. In this way, multiple representations of the single signal component are prevented.

In another embodiment of the invention the segmenting means is arranged to derive another signal segment from the input signal, the other signal segment being the first signal segment.
Longer than a signal segment, the audio encoder is configured to derive the encoded audio signal also based on the other signal segment, the audio encoder further comprising the encoded signal representing the other signal segment. And a subtracting means for deriving the first signal segment by subtracting the other synthesized audio signal from the signal representing the input signal. ing. Experiments have shown that it is advantageous to use consecutive segments with at least three different lengths, as the number of periods in the segment is neither too large nor too small.

In yet another embodiment of the invention, the audio encoder comprises a filter for deriving a filtered signal from the input signal and deriving the first signal segment from the filtered signal. To be configured. By filtering the input signal, it is possible to remove some signal components from the input signal, making the determination of the remaining signal components more reliable. Signal components that are no longer present in the first signal segment are present in the second (or other) signal segment and are determined there. As a result, a more complete representation of the output signal is obtained.

Yet another embodiment of the present invention is characterized in that the encoding means is arranged to represent amplitude on a psychoacoustic scale. As a result of using a psychoacoustic scale to represent the amplitude, the transmission channel is used more efficiently. This is because less symbols are needed to represent a signal with a given dynamic range. Such a psychoacoustic scale can be, for example, a logarithmic scale.

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION

  Hereinafter, the present invention will be described with reference to the drawings.

In the transmission system according to FIG. 1, the audio signal to be transmitted is supplied to the input of the transmitter 2. In the transmitter 2, the input signal is supplied to the audio encoder 4. In the audio encoder 4, the input signal is supplied to the first input terminal of the subtracter 12 and the input terminal of the analysis unit 8. The analysis unit 8 determines the amplitude, phase and frequency of the sinusoidal signal components present in its input signal.

The output end of the analysis unit 8 carrying an output signal representing the amplitude, phase and frequency of the sine signal component is connected to the input end of the combiner 10 and the input end of the multiplexer 16. The synthesizer generates a synthetic audio signal composed of a plurality of sinusoids based on the amplitude, phase and frequency input from the analysis unit 8.

The output of the synthesizer 10 for transmitting the synthesized audio signal is supplied to the second input terminal of the subtracter 12. The subtractor 12 subtracts the synthesized audio signal generated by the synthesizer 10 from the input signal.

The output signal of the subtractor 12 is supplied to the noise analyzer 14. This noise analyzer 14 determines the spectrum of the noise signal at its input. The expression information of the noise spectrum is supplied to the multiplexer 16. The multiplexer 16 combines the signals from the analysis unit 8 and the noise analyzer 14 into a composite signal (co
mbined signal).

The multiplexer 16 preferably uses a psychoacoustic model to determine which signal components determined by the analysis unit 8 are perceptually relevant. Only perceptually relevant signal components are transmitted. Psychoacoustic models that determine perceptually relevant signal components are commonly used in frequency domain encoders and are therefore known to those skilled in the art.

The output signal of the multiplexer 16 constitutes the output signal of the audio encoder 4. The output of the audio encoder 4 is connected to the input of a transmission unit 6, which produces a signal suitable for transmission via the transmission medium 3 to the receiver 24. The transmission unit 6 performs operations such as channel coding, interleaving and modulation.

The signal from the transmission medium 3 is supplied to the receiving unit 18 in the receiver 24. The receiving unit 18 performs operations such as demodulation, deinterleaving and channel decoding.

The output terminal of the receiving unit 18 is connected to the input terminal of the audio decoder 22. In the audio decoder 22, the signal from the receiving unit is supplied to a demultiplexer 20, which demultiplexes the first signal representing the sinusoidal signal component determined by the analysis unit 8 and the noise spectrum determined by the analyzer 14. And outputs a second signal that represents.

The first signal is supplied to a sinusoidal combiner 26, which derives a combined signal from the first signal. The combiner 26 is similar to the combiner 10 used in the encoder 4. The second signal is supplied to a noise synthesizer 28, which produces a noise signal having a spectrum defined by the second signal. This is done by performing an IFFT on the received spectrum, where random phases are assigned to the spectral components. The output signals of the sinusoidal synthesizer 26 and the noise synthesizer 28 are added by an adder 30 to obtain a duplicate of the input audio signal.

In the analysis unit (analyzer) 8 of FIG. 2, the input signal is supplied to the segmentation unit 42 and the input end of the low pass filter 30. The segmentation unit 42 selects a segment having 360 samples from the input signal. For a sample rate of 44.1 kHz of the input signal, this corresponds to an analysis period of 8.16 ms.

The output terminal of the low-pass filter 30 is connected to the input terminal of the decimator 32, and the decimator reduces the sample rate by a factor of 3. The low-pass filter 30 prevents false signals and has a cutoff frequency of 500 Hz. This cutoff frequency is much lower than that required for spurious protection, but is designed so that only signals with a small number of periods within the corresponding analysis window pass with little attenuation.

The output of the decimator 32 is connected to the input of the segmenting unit 40 and the input of the low pass filter 34. The segmentation unit 40 selects a segment having 360 samples from the output of the decimator 32. For a sample rate of 14.7 kHz (decreased), this corresponds to an analysis period of 24.5 ms.

The low pass filter 34 has a cutoff frequency of 165 Hz. The output of the low pass filter 34 is connected to the input of a decimator 36, which again reduces the sample rate by a factor of 3. The output of decimator 36 is connected to the input of segmentation unit 38, which selects the segment having 256 samples. For a sample rate of 4.9 kHz (reduced by a factor of 2),
This corresponds to an analysis cycle of 52.2 ms.

The output of segmentation unit 38 is provided to spectrum estimation unit 44,
The unit determines the spectral components by peak picking followed by a fine search in the Fourier domain. Several methods of estimating the sinusoidal component are well known to those skilled in the art of audio coding.

The output end of the spectrum estimation unit 44 is connected to the input end of the frequency selector 50. This frequency selector selects only frequency components within a well-defined range. In this example, the selector 50 selects only the frequency component having the maximum frequency of 133 Hz. Spectral components with higher frequencies are simply discarded. The corrector 52 corrects the amplitude and phase values of the selected signal component. This correction is needed to compensate for the amplitude and phase distortions introduced by the filter 34. Since the transfer function of this filter is known, the required correction factor can be easily determined.

The output of the corrector 52 is supplied to a synthesizer 54, which generates a synthetic voice signal based on the output signal of the corrector 52. The sample rate of the synthesized speech signal supplied by the synthesizer 54 corresponds to the sample rate at the output of the decimator 32. The synthesized audio signal provided by synthesizer 54 is subtracted from the output signal of segmenting unit 40 by subtractor 46. The combination of combiner 54 and subtractor 46 is part of the prevention means according to the invention. Therefore, the signal component determined by the estimation unit 44 and selected by the selection unit 50 is substantially removed from the output signal of the segmentation means 40.

The output signal of subtractor 46 is passed to spectrum estimation unit 55, which determines the spectral components in the output signal. Then the selection unit 5
6 selects only signal components with frequencies below 400 Hz.

The output terminals of the corrector 52 and the selector 56 are connected to the input terminal of a combiner 58. Combiner 58 combines the frequency estimates derived from the signal segments with different durations. In the finer time axis (shorter segments), we can find approximately the same frequencies as in the coarser time axis, so that the corresponding signal components can be represented by a single signal component. In this example, this combination is made when the frequencies differ by less than 10 ⁻³ rad.
The combiner 58 is also a part of the above-mentioned prevention means.

The output of combiner 58 is passed to corrector 62 to correct the amplitude and phase distortions of filter 30. The output signal of corrector 62 is provided to the input of combiner 60, which generates a composite audio signal based on the identified signal components. The synthesized audio signal generated by synthesizer 60 is subtracted from the output signal of segmenting unit 42 by subtractor 48. The combination of combiner 60 and subtractor 48 is part of the prevention means according to the invention. The output signal of subtractor 48 is passed to spectrum estimation unit 64, which determines the signal components in its input signal. These signal components, along with the output signal of corrector 62, are passed to combiner 68, which determines the representation of all sinusoids found in the input signal. The maximum number of sinusoids to be determined by the estimator 44 is chosen to be equal to 5, the maximum number of sinusoids to be determined together by the estimators 44 and 55 is 10, and the estimators 44, 55 and 64 are The total number of sinusoids to be determined by is determined to be equal to 60.

Since the output signals of the segmentation units 38, 40 and 42 have different lengths, the above analysis is also performed for different time axes. Preventing means for suppressing or preventing multiple representations of a single signal component are here the combiners 54 and 60, the subtractor 46.
And 48 and combiners 58 and 68. However, it is also conceivable that only the combination of the synthesizer and the subtractor mentioned above is used for the prevention means, or only the combiner is used for the prevention means.

In the diagram of FIG. 3, the signal segments used in the analyzer 8 are displayed. Graphs 70, 71 and 72 show the relevant signal segments at time T ₁ .

The graph 70 shows the segments available at the output of the segmentation unit 42 at time T ₁ . This segment has N = 360 samples. Graph 71 shows the segments available at the output of segmenting unit 40 at time T ₁ . This segment also has N = 360 samples.

The graph 72 shows the segments available at the output of the segmentation unit 38 at time T ₁ . In this case, the segment has M = 256 samples. From these graphs it is clear that different duration signal segments are used in the analysis.

Graphs 73, 74 and 75 show signal segments at a subsequent analysis time T ₂ . It can be seen that all segments are shifted to the right for the duration of the shortest segment. This is because the complete analysis is done in period T.
Graphs 76, 77 and 78 show the signal segment at time T _{3 which} is T later than T ₂ .

In the noise analyzer 14 of FIG. 4, the input signal is segmented by the segmenting means 80, 82.
And 84 input terminals. The segmenting means 80 uses the input signal 10
It is configured to derive a segment of 24 samples. The segmenting means 82 is also arranged to derive a segment of 512 samples from the input signal and the segmenting means 84 is arranged to derive a segment of 256 samples from the input signal.

The output of the segmenting means 80 is connected to the input of the FFT processor 86 for determining the frequency spectrum for the low frequency range. The FFT processor 86 is configured to perform a 1024 point FFT. The output of the segmenting means 82 is connected to the input of the FFT processor 90. The FFT processor 90 executes 512-point FFT. The output of the segmenting means 84 is connected to the input of the FFT processor 94. The FFT processor 94 performs a 256-point FFT.

To apply the psychoacoustic model in the multiplexer 16 of FIG. 1, it is desirable to represent the noise spectrum as noise power per ERB bin. To do so, the FFT bin values determined by FFT processors 86, 90 and 94 are converted to 18, 7 and 18 ERB bins by ERB converters 88, 92 and 96, respectively. Since all ERB bins cover different frequency ranges, ERB converters 88, 92 and 96 constitute suppression means that prevent multiple representations of signal components. FFT processors 86, 90 and 94 are complete FFTs
It is also conceivable to perform only a partial FFT that determines only the frequency bins necessary to determine the ERB bin corresponding to the FFT, without performing. In that case, the suppression means also include FFT processors 86, 90 and 94.

The ERB converters 88, 92 and 96 convert the value for each ERB bin to the corresponding ERB bin.
It is derived by adding the powers in the FFT bins that are within the range defined by The conversions to be performed by the ERB converter are:

[Equation 1] Can be written in the form of a matrix. In equation (1), Y (n) is each ER
The power in the B bin, where n represents the rank number of the ERB bin. P is a vector containing the power of the FFT bin as an element:

[Equation 2] Can be defined as In Expression (2), | X (k) | ² is the k-th FF
The power in the T bin. L is the number of points included in the FFT. The vector W (n) represents the overlap between the ERB bin and the FFT bin. If f ₁ represents the lower limit of the ERB bin and f ₂ represents the upper frequency of the ERB bin, the vector element W
(n, k) can be written as:

[Equation 3] In equation (3), b is the size of the FFT bin and is equal to f _s / L. Taking different values for n to get all ERB bins:

[Equation 4] The matrix multiplication according to is obtained. The power in the ERB bin is
Noise analyzer 1 for use with a psychoacoustic model used in
4 additional outputs.

The noise synthesizer 28 inverse transforms W of W to obtain the FFT bin from the ERB bin.
Need ~. This reciprocal W ~ can be obtained in the same way as W was determined. This reciprocal W ~ is:

[Equation 5] Can be calculated by The 43 ERB power values are passed to the fitting means 98, which performs the fitting of the cubic polynomial to the 43 ERB power values. Therefore, the estimated powers are aligned in time (these are estimated at different analysis segment sizes). As a result of this fitting process, the data is reduced from 43 coefficients to 4 coefficients. Prior to performing the fit, the amplitudes in the ERB bins are converted to values on a psychoacoustic scale, such as a logarithmic scale or an approximation thereof.

In the combiner 28, the 43 ERB power values are calculated by the cubic polynomial defined by the four coefficients. The combination is done on different timeframes for different groups of ERB powers, just as done in the above analysis.

Although the present invention has been described above with reference to preferred embodiments, it should be understood that these are not limiting examples. Thus, various modifications will be apparent to those skilled in the art without departing from the scope of the invention as claimed.

As an example, it is also possible that consecutive signal segments partially overlap, although they do not overlap in the above embodiments. Furthermore, the separate preventive measures disclosed in the above embodiments do not necessarily have to be present in combination and may be used separately.

In sinusoidal audio encoders, it is known to use different time axes to analyze different parts of the frequency spectrum. In conventional encoders, subband filtering is used to split the input signal into multiple subbands.

By splitting the input signal into sub-bands, it is possible that the signal components at the boundary of two sub-bands will be an expression in both sub-band signals. This dual representation of signal components leads to some problems in encoding these components. According to the invention, prevention means (46, for preventing the signal component from having multiple representations)
48, 58, 68; 88, 92, 96).

[Brief description of drawings]

[Figure 1]   FIG. 1 shows a transmission system in which the present invention can be used.

2 shows an analysis unit 8 for sinusoids according to the invention to be used in the transmission system of FIG.

[Figure 3]   FIG. 3 shows the signal segments used in the analysis unit 8 of FIG.

FIG. 4 shows a noise analyzer 14 according to the invention to be used in the transmission system of FIG.

[Explanation of symbols]

  2 ... transmitter   3 ... Transmission medium   4 ... Audio encoder   6 ... Transmission unit   8 ... Analysis unit 10 ... Synthesizer 12 ... Subtractor 14 ... Noise analyzer 16 ... Multiplexer 18 ... Receiving unit 20 ... Demultiplexer 22 ... Audio decoder 24 ... Receiver 26 ... Synthesizer 28 ... Noise synthesizer 30, 34 ... Low-pass filter 32, 36 ... Decimator 38, 40, 42 ... Segmentation unit 44, 55, 64 ... Spectrum estimation unit 46, 48 ... Subtractor 54, 60 ... Synthesizer 58, 68 ... Combiner 88, 92, 96 ... ERB converter

─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Omen Arnordas Doubler             I             Netherlands 5656 aer ind             Fenprof Holstraan 6 (72) Inventor Denblinker Albertus C             Netherlands 5656 aer ind             Fenprof Holstraan 6 F term (reference) 5D045 DA11                 5K041 AA01 CC01 DD01 FF01 HH12

Claims (15)

[Claims] 1. A transmitter having an audio encoder, said audio encoder having segmenting means for deriving at least a first signal segment and a second signal segment from an input signal representative of an audio signal, said first encoder The signal segment is longer than the second signal segment, the audio encoder has coding means for deriving a coded audio signal from the first and second signal segments, and the transmitter has the coded audio. Transmitting means for transmitting a signal to a receiver via a transmission medium, the receiver having receiving means for receiving the encoded audio signal from the transmission medium, and the encoded audio signal In a transmission system further comprising an audio decoder for deriving from the encoded audio signal The transmission system is characterized in that the encoding means has a prevention means for preventing a multiple representation of a single signal component from occurring in the encoded audio signal. 2. The transmission system according to claim 1, wherein the preventing unit is
Combining means for deriving a composite audio signal from a portion of the encoded audio signal representing the first signal segment, and subtracting the composite audio signal from a signal representing the input signal to generate the second signal segment. And a subtraction unit for deriving the transmission system. 3. The transmission system according to claim 2, wherein the segmenting means is configured to derive another signal segment from the input signal, the other signal segment being longer than the first signal segment, The audio encoder is configured to derive the encoded audio signal also based on the other signal segment, and the preventing means is a portion of the encoded audio signal representing the other signal segment. And another synthesizing means for deriving another synthetic signal from
And a subtraction means for deriving the first signal segment by subtracting the other synthesized audio signal from a signal representing the input signal. 4. The transmission system according to claim 1, wherein the audio encoder has a filter that derives a filtered signal from the input signal, and derives the first signal segment from the filtered signal. A transmission system characterized by being configured to: 5. The transmission system according to claim 4, wherein the filter is
A transmission system comprising decimating means for obtaining said first signal segment at a reduced sample rate. 6. Transmission system according to claim 1, characterized in that the coding means are arranged to represent the amplitude on a psychoacoustic scale. 7. A transmitter having an audio encoder, said audio encoder having segmenting means for deriving at least a first signal segment and a second signal segment from an input signal representing the audio signal, First
The signal segment is longer than the second signal segment, the audio encoder has coding means for deriving a coded audio signal from the first and second signal segments, and the transmitter has the coded audio. A transmitter, comprising transmitting means for transmitting a signal, wherein said encoding means comprises preventing means for preventing a multiple representation of a single signal component from occurring in said encoded audio signal. 8. The transmitter according to claim 7, wherein the preventing means comprises a combining means for deriving a combined audio signal from a portion of the encoded audio signal representing the first signal segment, and the input. Subtracting means for deriving the second signal segment by subtracting the synthesized audio signal from a signal representing the signal. 9. An audio encoder having segmenting means for deriving at least a first signal segment and a second signal segment from an input signal representing an audio signal, the first signal segment being longer than the second signal segment. , The audio encoder has coding means for deriving a coded audio signal from the first and second signal segments, the coding means multiplexing a single signal component into the coded audio signal. An audio encoder having a preventing means for preventing an expression from occurring. 10. The audio encoder according to claim 9, wherein the preventing means comprises a synthesizing means for deriving a synthesized audio signal from a part of the encoded audio signal representing the first signal segment, and the input. An audio encoder, the subtraction means deriving the second signal segment by subtracting the synthesized audio signal from a signal representing the signal. 11. Derivation of at least a first signal segment and a second signal segment from an input signal representing an audio signal, said first signal segment being longer than said second signal segment and encoded. An audio coding method comprising the step of deriving an audio signal from the first and second signal segments, the method generating a multiple representation of a single signal component in the coded audio signal. An audio encoding method comprising the step of preventing 12. The method of claim 11, wherein the method represents a composite audio signal from a portion of the encoded audio signal representing the first signal segment, the method representing the input signal. Deriving the second signal segment by subtracting the synthesized audio signal from a signal. 13. A computer program enabling a processor to carry out the method of claim 11. 14. A tangible medium having the computer program of claim 13. 15. A signal carrying a computer program as claimed in claim 13.